Module: Mesa Branch: main Commit: d2ffcb609252a4f2892df41649c84f30ce4c940b URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=d2ffcb609252a4f2892df41649c84f30ce4c940b
Author: Faith Ekstrand <[email protected]> Date: Tue Dec 5 20:05:07 2023 -0600 nir: Lower [su]dot_4x8_[ui]add_sat to [su]dot_4x8_[ui]add Since nir_opt_algebraic runs on its own results, if the driver doesn't have [su]dot_4x8_[ui]add then the [su]dot_4x8_[ui]add lowering rules will kick in and lower that to what we had originally. Reviewed-by: Jesse Natalie <[email protected]> Reviewed-by: Ian Romanick <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26533> --- src/compiler/nir/nir_opt_algebraic.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py index d16bdb84e5f..6cb19b53343 100644 --- a/src/compiler/nir/nir_opt_algebraic.py +++ b/src/compiler/nir/nir_opt_algebraic.py @@ -321,15 +321,15 @@ optimizations.extend([ # overflowing. 0x100000000 - 0x3f804 = 0xfffc07fc. If c is a constant # that is less than 0xfffc07fc, then the result cannot overflow ever. (('udot_4x8_uadd_sat', a, b, '#c(is_ult_0xfffc07fc)'), ('udot_4x8_uadd', a, b, c)), - (('udot_4x8_uadd_sat', a, b, c), ('uadd_sat', udot_4x8_a_b, c), '!options->has_udot_4x8_sat'), + (('udot_4x8_uadd_sat', a, b, c), ('uadd_sat', ('udot_4x8_uadd', a, b, 0), c), '!options->has_udot_4x8_sat'), # For the signed dot-product, the largest positive value is 4*(-128*-128) = # 0x10000, and the largest negative value is 4*(-128*127) = -0xfe00. We # don't have to worry about that intermediate result overflowing or # underflowing. - (('sdot_4x8_iadd_sat', a, b, c), ('iadd_sat', sdot_4x8_a_b, c), '!options->has_sdot_4x8_sat'), + (('sdot_4x8_iadd_sat', a, b, c), ('iadd_sat', ('sdot_4x8_iadd', a, b, 0), c), '!options->has_sdot_4x8_sat'), - (('sudot_4x8_iadd_sat', a, b, c), ('iadd_sat', sudot_4x8_a_b, c), '!options->has_sudot_4x8_sat'), + (('sudot_4x8_iadd_sat', a, b, c), ('iadd_sat', ('sudot_4x8_iadd', a, b, 0), c), '!options->has_sudot_4x8_sat'), (('udot_2x16_uadd_sat', a, b, c), ('uadd_sat', udot_2x16_a_b, c), '!options->has_dot_2x16'), (('sdot_2x16_iadd_sat', a, b, c), ('iadd_sat', sdot_2x16_a_b, c), '!options->has_dot_2x16'),
